81 replies to this topic

[Knight Captain Morgan]

Eatit, on 26 April 2019 - 08:43 AM, said:

I like your chart McGoat but the number of tier 1 players on there is pretty small. How will they get a decent wait time with such a low population?

A good matchmaker could put tier 1's and 5's in the same match by evenly distributing them instead of lumping all of T1's on 1 side.

Say there's 24 players, 6 per weight class. Matchmaker ranks all players within each weight class then starts building the teams.

Blue team would get rank 1, 4, and 6 of the assaults and lights while red team gets ranks 2,3, and 5 of the assaults and lights.

Blue team would get rank 2,3, and 5 of the heavies and mediums while red team gets ranks 1,4, and 6 of the heavies and mediums.

It doesn't have to be an even 6 per weigh class just so long as it tries to give best and worst within each weight class on the same team to balance things out. Say 10 assaults, 6 heavies, 2 mediums, and 6 lights in a match.

Blue team gets ranks 1, 4, 6, 8, and 10 of the assaults, 2,3, and 5 of the heavies, rank 2 of the mediums, and 1, 4, and 6 of the lights. etc

edit - regardless of what they do or don't do, they've only got 136 more days.

Edited by Andres Gomez, 27 April 2019 - 07:37 AM.

[vandalhooch]

PhoenixFire55, on 26 April 2019 - 03:12 AM, said:

Well, I got bad news for you and him ... In any kind of 50/50 experiment, be it a coin toss or a MM setting you up for a guaranteed win/loss, there is a nearly 100% probability that the amount of same events happening in a row will be equal to square root of the number of experiments done. I.e. you play 100 matches, be prepared to have a 10 match long win/loss series with near 100% chance.

Your math based arguments would be so much more persuasive if you could do any actual math.

With a .5 win rate, a 10-loss streak has a .000097 probability of occurring during any particular 10 match series. In any 100 match series there are 90 10-match series that take place. That means there is only an 8.7% chance that a 10-match loss streak occurs during that 100-match series (90 x .000097).

I would argue that 8.7% is not quite the "near 100% chance" you claim.

Edited by vandalhooch, 27 April 2019 - 10:07 AM.

[PhoenixFire55]

vandalhooch, on 27 April 2019 - 10:05 AM, said:

Your math based arguments would be so much more persuasive if you could do any actual math.

With a .5 win rate, a 10-loss streak has a .000097 probability of occurring during any particular 10 match series. In any 100 match series there are 90 10-match series that take place. That means there is only an 8.7% chance that a 10-match loss streak occurs during that 100-match series (90 x .000097).

I would argue that 8.7% is not quite the "near 100% chance" you claim.

Yeah well, it helps to check your zeroes ... You got an extra one, its .00097 actually, and I got one missing, its obviously 96.7% in 1000 matches.

[vandalhooch]

PhoenixFire55, on 27 April 2019 - 10:23 AM, said:

Yeah well, it helps to check your zeroes ... You got an extra one, its .00097 actually, and I got one missing, its obviously 96.7% in 1000 matches.

1 - Yep, I had a typo in my post but not in my calculator.

2 - So a near guarantee in every 1000 matches versus every 100 matches . . . seems your original argument is not nearly as effective as you thought.

[Jman5]

Nesutizale, on 23 April 2019 - 01:09 PM, said:

A nice video I stumbled upon https://youtu.be/NXD...j_Qk?t=2483
Starting around 41:23 they start talking about ELO and W/L ratings.

So it seams that the idea of the PSR rating is indeed better for 12v12 matchups with random people then an ELO system.

Also it makes me wondern, how is the new faction play system working? Is it an PSR or ELO system or a mix?

PS If you see the entire video, there is a lot in it that makes me think about MWO and why its not working so well compared to other games.

Do we have a noob tube? Whats with the learning curve and stuff...

Well I decided to slug through the entire video. Those guys are smart, but I do disagree with them on a couple of points.

Assertion: Elo doesn't work in team base games because you could get stuck with bad teammates and lose through no fault of your own.

While it's certainly true that you will sometimes get saddled with duds, it also doesn't matter. Over a period of time, random team mode (solo queue) will give every player an equal number of good and bad teams. Their sports team analogy only works if we're speaking about full premade teams. Frankly, if you just play with that 1 group, your team is going to wind up with a good Elo.

MWO's Elo could have worked if they had separated solo and group queue. PGI was even pretty clever about it by splitting it up by weight class.

----

One thing I was pleasantly surprised to see them bring up is the idea of handicaps. That is something I wish we would see more in team based multiplayer games because it can make the game more enjoyable for casual players and create more competitive environments for higher skilled players.

What I would really like to see is a second wave light/medium mech for the first two or three guys to die on each team. It could be a stock mech, just something to let them get a little more play time. It would also make the games a little longer which I think is great.

[PhoenixFire55]

vandalhooch, on 27 April 2019 - 10:45 AM, said:

1 - Yep, I had a typo in my post but not in my calculator.

2 - So a near guarantee in every 1000 matches versus every 100 matches . . . seems your original argument is not nearly as effective as you thought.

Well, I couldn't really bother to waste time thinking this through, but since you insist ...
Your way of calculating the probability is actually quite wrong in the first place. You don't just multiply the probability of it happening by the number of possible series. Gotta calculate it properly and do the series of conditional probabilities ...

P = p + p*(1-p) + p*(1-p)^2 + ... + p*(1-p)^n = p*(1-(1-p)^n)/(1-(1-p)) = 1-(1-p)^n

Quite different from p*n isn't it? ...

The whole point was to show that the seires of same event happening in a row, weren't just a rare occurence, but rather in fact perfectly common in a symmetrical experiment. I can easily change my argument from 10 in a row, to 5 in a row happening inside a 100 matches long series, and it will happen with 95% probability, i.e. nearly guaranteed.

[Snowhawk]

Sometimes I have the Feeling that the matchmaker tries to put the same mechs on one side.

E.g. I saw one Team with 3 Executioners (3? they are really seldom on the Battlefield). Or in another match we had 3 Cyclopses. Or 2 Mad Cat MKII with 2 Blood Asps…..

[LordNothing]

as for balance waste of time. unless we have another glaring kodiak 3 situation do anything other than balance tweaks. if you arent makeing mechanics changes or make the weapons more interesting it practically any other way, dont do it.

the problem with the match maker is that its not what it appears to be. its a noob shield disguised as a skill rating. all it does is artificially fluff everyone's ego so they can pat them selves on the back and call themselves elite because of their t1 rating, all while having to put up with terrible games because you are rated higher than you should be. hell im t2 at best, i wish the exp bar would figure that out. either make the skill changes neutral, or make everyone bubble towards tier 3. unless you are so good you can fight bubble down or so bad you never bubble up, you end up in the place that lines up with the center of the bell curve where the bulk of the population is.

[Sjorpha]

justcallme A S H, on 27 April 2019 - 04:13 AM, said:

Yeah it's really not a hard concept to grasp as I see it.

Many seem to be unable to understand it however or simply don't understand how it works right now.

The entire point is exactly as I said a couple pages back. 190 Avg Match score and you make it into T1? That's the problem.

Yeah, and there are some other advantages to only balancing total team skill rather than trying to separate matches by tier.

A problem with separating matches by tier is that it inflates the results of bad players. For example a tier 5 match will have about the same avg. matchscore as a tier 1 match, so the ranking system will think the best/worst players in the tier 5 match and the best/worst players in the tier 1 match performed the same and adjust their rankings by the same amount, so separating matches by tier leads to players, especially average skill players, bouncing up and down in the ranking rather than ever settling at their correct rating. It also gives players wrong impressions about their own skills, someone in the tier 5 match will be scoring 1000+ damage, compare to similar scores posted by tier 1 players and incorrectly believe their performance is comparable.

If you scrap the idea of separation by tiers and just match the total skill of the teams, you will get a much more representative performance curve for the ranking system to work with. The "downside" side of that would of course be that bad players would actually get more consistently bad results which might be a bit demoralising, on the other hand they would have more skilled teammates to learn from and more realistic stats to improve upon, as well as a closer to 50/50 winrate. The best players would also have a much more even winrate in such a system.

Edited by Sjorpha, 29 April 2019 - 06:50 AM.

[Wil McCullough]

Sjorpha, on 29 April 2019 - 06:48 AM, said:

Yeah, and there are some other advantages to only balancing total team skill rather than trying to separate matches by tier.

A problem with separating matches by tier is that it inflates the results of bad players. For example a tier 5 match will have about the same avg. matchscore as a tier 1 match, so the ranking system will think the best/worst players in the tier 5 match and the best/worst players in the tier 1 match performed the same and adjust their rankings by the same amount, so separating matches by tier leads to players, especially average skill players, bouncing up and down in the ranking rather than ever settling at their correct rating. It also gives players wrong impressions about their own skills, someone in the tier 5 match will be scoring 1000+ damage, compare to similar scores posted by tier 1 players and incorrectly believe their performance is comparable.

If you scrap the idea of separation by tiers and just match the total skill of the teams, you will get a much more representative performance curve for the ranking system to work with. The "downside" side of that would of course be that bad players would actually get more consistently bad results which might be a bit demoralising, on the other hand they would have more skilled teammates to learn from and more realistic stats to improve upon, as well as a closer to 50/50 winrate. The best players would also have a much more even winrate in such a system.

A lot of what you mention are actually good things lol. I don't see why a tier 5 player isn't allowed to pull 1000 damage games when he's playing with other tier 5s. If he is able to consistently pull 1000 damage, he'll rank up to tier 4 and then 3 or whatever level he's actually playing at. Eventually, he'll get slotted into the tier he belongs and he won't be consistently getting high damage games anymore.

This means that every time every player starts a match, there is a trust in the matchmaker to make a game where he has a good chance of participating and contributing to wins and losses. That is great for player retention. No one wants to start out and get utterly goomba stomped while higher tier players in the team berate them for being burdens. It's better to start in the kiddie pool when learning how to swim.

Another good thing is that when everyone hovers around 50-50 wlr, the way to tell a good player apart from a bad/newer one isn't through wlr and kdr. It's through tier. That makes tier have a value. Right now it doesn't. A tier 1 pilot could be as tater as a tier 5. He just played longer and didn't learn anything.

Some cons though. When players are put into brackets of equal skill, there is much more personal responsibility for winning or losing. Which means incidents of players berating their teammates for stupid moves or mistakes.will increase. Because those will actually cost a team the game in higher tier play.

Second problem. The time it takes to get matches will also increase by a.lot, especially for tiers at the end of the curves like 1 and 5. The issue mwo's matchmaker has to deal with is that it has to not only get together 24 players of the same.tier, it has to factor in the mechs they choose to drop in. Mwo's low pop also means finding matches in this case woild be a nightmare.

[MrMadguy]

PhoenixFire55, on 29 April 2019 - 12:17 AM, said:

Well, I couldn't really bother to waste time thinking this through, but since you insist ...
Your way of calculating the probability is actually quite wrong in the first place. You don't just multiply the probability of it happening by the number of possible series. Gotta calculate it properly and do the series of conditional probabilities ...

P = p + p*(1-p) + p*(1-p)^2 + ... + p*(1-p)^n = p*(1-(1-p)^n)/(1-(1-p)) = 1-(1-p)^n

Quite different from p*n isn't it? ...

The whole point was to show that the seires of same event happening in a row, weren't just a rare occurence, but rather in fact perfectly common in a symmetrical experiment. I can easily change my argument from 10 in a row, to 5 in a row happening inside a 100 matches long series, and it will happen with 95% probability, i.e. nearly guaranteed.

There are 2^N ways to play N matches. And there N - M + 1 ways to lose M times in a row. In case of 50/50 chance, each of 2^N ways has 0.5^N chance to happen. And therefore each of N - M + 1 ways has (N - M + 1) * (0.5^N) or (N - M + 1) / (2^N) chance to happen. But at the same time there are 2^(N - M) ways, each of N - M + 1 way can happen, cuz we don't care about other matches, so they're the same for us. And therefore overall chance is (N - M + 1) * 2^(N - M) / 2^N = (N - M + 1) / 2^M = (N - M + 1) * (0.5^M).

In case of N = 100 and M = 10 it's 91 / 1024 = 0.089 or 8.9%

And it's the same, as 91*(0.5)^10. Why 91? Because there are 2 ways to lose 2 times in a row in 3 matches, not 1. I.e. formula is N - M + 1, not just N - M.

Edited by MrMadguy, 29 April 2019 - 10:27 PM.

[PhoenixFire55]

MrMadguy, on 29 April 2019 - 10:22 PM, said:

There are 2^N ways to play N matches. And there N - M + 1 ways to lose M times in a row. In case of 50/50 chance, each of 2^N ways has 0.5^N chance to happen. And therefore each of N - M + 1 ways has (N - M + 1) * (0.5^N) or (N - M + 1) / (2^N) chance to happen. But at the same time there are 2^(N - M) ways, each of N - M + 1 way can happen, cuz we don't care about other matches, so they're the same for us. And therefore overall chance is (N - M + 1) * 2^(N - M) / 2^N = (N - M + 1) / 2^M = (N - M + 1) * (0.5^M).

In case of N = 100 and M = 10 it's 91 / 1024 = 0.089 or 8.9%

And it's the same, as 91*(0.5)^10. Why 91? Because there are 2 ways to lose 2 times in a row in 3 matches, not 1. I.e. formula is N - M + 1, not just N - M.

Lets take your (N - M + 1) * (0.5^M) proability formula and subject it to a simple example, N=2, M=1, i.e. the probability to lose one match in a row in a series of two matches. Formula gives probabilty = 1, which quite frankly is false, because possible outcomes are W+W, W+L, L+W, L+L, i.e. only two out of four meet the requirement (three if we don't care about a series of exact length but not longer) thus its 0.5 probability (or 0.75).

Now try same formula with N=3, M=2. Formula gives 0.5 probability. Actual outcomes are WWW, WWL, WLW, WLL, LWW, LWL, LLW, LLL, two in a row happen in two of them (three with longer series), i.e. the actual answer is 0.25 (or 0.375).

For N=4, M=2, it gives probability = 0.75. While only 5 out of 16 outcomes (8 in case of longer series) are two losses in a row, i.e. its 0.31 (or 0.5). That formula just dosen't work.

[Sjorpha]

Wil McCullough, on 29 April 2019 - 05:08 PM, said:

A lot of what you mention are actually good things lol. I don't see why a tier 5 player isn't allowed to pull 1000 damage games when he's playing with other tier 5s. If he is able to consistently pull 1000 damage, he'll rank up to tier 4 and then 3 or whatever level he's actually playing at. Eventually, he'll get slotted into the tier he belongs and he won't be consistently getting high damage games anymore.

This means that every time every player starts a match, there is a trust in the matchmaker to make a game where he has a good chance of participating and contributing to wins and losses. That is great for player retention. No one wants to start out and get utterly goomba stomped while higher tier players in the team berate them for being burdens. It's better to start in the kiddie pool when learning how to swim.

Another good thing is that when everyone hovers around 50-50 wlr, the way to tell a good player apart from a bad/newer one isn't through wlr and kdr. It's through tier. That makes tier have a value. Right now it doesn't. A tier 1 pilot could be as tater as a tier 5. He just played longer and didn't learn anything.

Some cons though. When players are put into brackets of equal skill, there is much more personal responsibility for winning or losing. Which means incidents of players berating their teammates for stupid moves or mistakes.will increase. Because those will actually cost a team the game in higher tier play.

Second problem. The time it takes to get matches will also increase by a.lot, especially for tiers at the end of the curves like 1 and 5. The issue mwo's matchmaker has to deal with is that it has to not only get together 24 players of the same.tier, it has to factor in the mechs they choose to drop in. Mwo's low pop also means finding matches in this case woild be a nightmare.

Yeah, there are definitely pros and cons to the different ways of matchmaking. You do lose the "kiddie pool" for new players with the approach I'm suggesting which can be a problem, the tier system if it worked would definitely have the benefit of allowing low tier players to contribute more and feel better about it.

However even if you scrap the tiering model you could still do some kind of kiddie pool just for new players, let's say the first 100 games are played in a separate queue with only new players, then you enter the regular matchmaker.

The downside to having a "kiddie pool" though is that there won't be veterans to ask advice and learn from, and to me that's a pretty significant con. It's also a con of a tiered matchmaker, if you never see the best players because you play in a lower tier you also won't see and learn what the strongest builds and strategies are, and if your goal is to git gud that will slow you down.

But none of those those pros and cons matter if the system doesn't work. We've never had the tier system successfully separating all the tiers reliably, so we aren't really seeing the potential pros of the tiered system. It does manage to keep a short term kiddie pool for new players (evident if you make an alt account) so there's that. But I still say it's better to just accept that there isn't the population for tiering, and the matchmaker should be downscaled to just try and get 2 teams of equal total skill rating for each match. It would have some cons to it (and some pros) but it would at least function with both low and high population.

Getting a better functioning PSR/ELO/rating formula that doesn't have an upwards bias and doesn't cap out so can understand the skill difference between the good and the best players is a separate issue, and a big one. It could be a zero sum bell curve type system or it could be a limitless system of some kind (for example ELO type systems has no upper limits), there are many kinds of systems that could work but the important thing is it has to retain enough granularity to produce rankings all the way from the worst to the best player that the matchmaker can work with.

IMO this capping out at the top is the main problem with current PSR. Upwards bias could in theory keep working indefinitely if there was no upper limit to PSR. You'd get ridiculously high numbers over time for the top players but at least it would still know that Vorteex is a lot better than I am.

[Wil McCullough]

Hmm. I don't think ingame mid-match is the place for new players to ask for guidance. That should be done on other communication streams. Like twitch, the forums and whatnot. The danger of that is they may get ambushed by REALLY BAD ADVICE from players like the one whose name means "cunning".

I actually don't know why the tiers have always been smooshed together even when player numbers were higher. Pgi is also extremely opaque on that issue. They open and close valves behind the scenes without informing anyone or they may have shut off the matchmaking system completely for all anyone knows. I find it extremely hard to.conclude that it's a matchmaker problem because of all these shenanigans. The matchmaker hasn't been given an opportunity to work due to all these cockblock tinkering and a ridiculous ranking system that forces everyone upward regardless of skill level.

I can't even wrap my head around the reasons behind that decision. Was it to ensure almost every player felt a sense of progression? Or improvement? Did they actually think bad pilots within the system would be tempted to play more matches ans get to tier 1? It's warped beyond belief.

[MrMadguy]

PhoenixFire55, on 30 April 2019 - 12:36 AM, said:

Lets take your (N - M + 1) * (0.5^M) proability formula and subject it to a simple example, N=2, M=1, i.e. the probability to lose one match in a row in a series of two matches. Formula gives probabilty = 1, which quite frankly is false, because possible outcomes are W+W, W+L, L+W, L+L, i.e. only two out of four meet the requirement (three if we don't care about a series of exact length but not longer) thus its 0.5 probability (or 0.75).

Now try same formula with N=3, M=2. Formula gives 0.5 probability. Actual outcomes are WWW, WWL, WLW, WLL, LWW, LWL, LLW, LLL, two in a row happen in two of them (three with longer series), i.e. the actual answer is 0.25 (or 0.375).

For N=4, M=2, it gives probability = 0.75. While only 5 out of 16 outcomes (8 in case of longer series) are two losses in a row, i.e. its 0.31 (or 0.5). That formula just dosen't work.

Yeah, there are overlaps in 2^(N-M) formula. I need to figure out, how to eliminate them.

[PhoenixFire55]

MrMadguy, on 30 April 2019 - 02:33 AM, said:

Yeah, there are overlaps in 2^(N-M) formula. I need to figure out, how to eliminate them.

As much as I admire the dedication, its not really worth it ... it doesn't change the initial point.

[vandalhooch]

PhoenixFire55, on 29 April 2019 - 12:17 AM, said:

Well, I couldn't really bother to waste time thinking this through, but since you insist ...
Your way of calculating the probability is actually quite wrong in the first place. You don't just multiply the probability of it happening by the number of possible series. Gotta calculate it properly and do the series of conditional probabilities ...

P = p + p*(1-p) + p*(1-p)^2 + ... + p*(1-p)^n = p*(1-(1-p)^n)/(1-(1-p)) = 1-(1-p)^n

Quite different from p*n isn't it? ...

You know that makes your result even less likely to occur, right? It puts the likelihood around 1 in 2046.

If you could do the math originally, why did you claim that they were "guaranteed" to see a ten loss streak in 100 matches?

Quote

The whole point was to show that the seires of same event happening in a row, weren't just a rare occurence, but rather in fact perfectly common in a symmetrical experiment.

I wasn't arguing against that. I was pointing out that your "example" wasn't actually valid.

Quote

I can easily change my argument from 10 in a row, to 5 in a row happening inside a 100 matches long series, and it will happen with 95% probability, i.e. nearly guaranteed.

And a 5-loss streak has the same psychologically impact on a player as a 10-loss streak?

[PhoenixFire55]

vandalhooch, on 05 May 2019 - 02:20 PM, said:

If you could do the math originally, why did you claim that they were "guaranteed" to see a ten loss streak in 100 matches?

Simple really, couldn't care to double check, since it has little to do with an actual point I was making.

vandalhooch, on 05 May 2019 - 02:20 PM, said:

I wasn't arguing against that. I was pointing out that your "example" wasn't actually valid.

Yeah well, you aren't the one to talk, given your math wasn't valid either.

vandalhooch, on 05 May 2019 - 02:20 PM, said:

And a 5-loss streak has the same psychologically impact on a player as a 10-loss streak?

Heh. Even a 2-loss streak seemingly makes some people throw a fit about it. You keep seeing folks complaining about how "they lose 10-20 in a row" and "only win 1 out of 10 matches at best", yet then you check their stats and they all have a perfectly normal ~1.0 W/L ratio. Its psycological really ... a 5-loss streak that is very likely to happen becomes a 10-loss streak in their mind and then a 20-loss streak the next time they retell the tale on the forum. So ... yes, it has.

[Sjorpha]

Not doing or arguing the math as people are already doing that, but the main point is that a series of for example 10 losses in a row isn't a surprising event in a random sequence of wins and losses. It's bound to happen to some players every now and then given there is thousands of players and some of them have played thousands of games.

Also worth pointing out is that any specific series of random numbers is equally likely. So if you are throwing a d6 10 times you are exactly as likely to get 6666666666 as you are to get for example the exact sequence 1362531262. Both of those series have the same low probability of happening, most people just have a stronger psychological response to 6666666666 than to 1362531262. Point here being that 10 losses in a row is exactly as likely as any other specific series of 10 with mixed wins and losses, each specific sequence is unlikely but you always get one and every once in a while someone will get the 10 losses.

Of course that all assumes a 50/50 probability of winning and losing, in reality your chance of winning and losing each match is affected by how good you are and there are players with all the way up to an incredible 80% win chance per game (W/L of 4+)

Edited by Sjorpha, 06 May 2019 - 06:01 AM.

[Aidan Crenshaw]

If you're throwing a 7 on a D6, you're cheating!